Jensen Huang wants everyone to calm down. Speaking at Y Combinator's Startup School, the Nvidia chief executive dismissed the "white-collar bloodbath" warnings that have shadowed the AI era and offered a tidy distinction in their place. AI eliminates tasks, he argued, not jobs. "Every single job will change, and there will be a whole bunch of new jobs," he said. "The narrative of AI destroying jobs is exactly backward." It is a genuinely useful frame. It also papers over the harder question of what happens to the people whose careers were built on the tasks now being automated away.
Huang's logic is that a job is a bundle of tasks wrapped around a purpose, and automating some of the tasks frees people to pursue more of the purpose. His examples are real. Radiologists were supposed to be replaced by image-reading algorithms years ago; instead, with the reading sped up, hospitals clear their patient backlogs faster and radiology roles keep growing. Software engineers now spend less time typing code and more time directing tools like Claude Code and Codex, and engineering headcount is rising, not falling. "If we can automate away the task of programming," Huang said, "we could hire more software engineers to do more things."
The trouble is that the same speech contained its own counterexample. Huang noted that the phone-and-database work of customer service "will be automated away," and just last week Uber cut 10 percent of its customer-service staff, citing exactly that technology. Those are not tasks migrating to higher purposes; they are jobs ending. His nod to the legal startup Harvey, which he says is making paralegal work grow "like crazy," sits awkwardly against Bureau of Labor Statistics projections of little or no change in that role through 2034. Even Anthropic's Dario Amodei, who last year warned AI could wipe out half of entry-level white-collar jobs, has softened toward the productivity story while still calling some job loss "intrinsic" to the technology.
What the tasks-versus-jobs debate tends to skip is where expertise comes from. A Swedish company called Lightbringer offers a preview. It is what its co-founder Dominic Davies calls a "full-stack service firm": it owns the AI platform and employs the patent attorneys, contracting directly for the outcome rather than licensing software to a law firm. Its system breaks patent work into small decisions, identifying claim components, comparing each against prior art, testing novelty, the very sequence junior attorneys once learned by grinding through it. "This is going to come for everybody," Davies says, not just the junior ranks.
That reversal is the part worth dwelling on. Professional judgment has traditionally been built by doing the boring, repetitive junior work, making the first error and learning why it failed. Automate that away, and reviewing an AI-produced draft is not the same teacher. If the machine handles the apprenticeship-grade tasks, where does the next generation of reviewers acquire the judgment to review? Lightbringer's own metrics, more than 200 clients and 300 percent quarterly revenue growth, measure commercial traction, not whether its patents survive examination or litigation. The proof of the model, as one account of the company put it, will come the day a filing fails and someone has to decide who is responsible.
Huang is probably right that aggregate employment will grow, and the doom forecasts have so far overshot. But "tasks, not jobs" is a description of the average, and averages hide the transition. The radiologist keeps working; the person who would have become a radiologist by reading ten thousand routine scans may find that ladder missing its lowest rungs. The optimism and the anxiety are not really in conflict. They are describing the same shift, from different ends of a career.